Exploiting human steering models for path prediction

The ability to predict the path of a moving human is a crucial element in a wide range of applications, including video surveillance, assisted living environments (smart homes), and simulation environments. Two tasks, tracking (finding the user's current location) and goal prediction (identifying the final destination) are particularly relevant to many problems. Although standard path planning approaches can be used to predict human behavior at a macroscopic level, they do not accurately model human path preferences. In this paper, we demonstrate an approach for path prediction based on a model of visually-guided steering that has been validated on human obstacle avoidance data. By basing our path prediction on egocentric features that are known to affect human steering preferences, we can improve on strictly geometric models such as Voronoi diagrams. Our approach outperforms standard motion models in a particle-filter tracker and can also be used to discriminate between multiple user destinations.